Sign In

Remember Me

Singularity Summit – Itamar Arel on Artificial General Intelligence

Itamar Arel’s talk focused on AGI (Artificial General Intelligence), where he described a two pronged path toward achieving AGI that might take us years instead of decades. He believes that we already have the pieces of the puzzle to build human level AI system, that it’s really a question of creating a properly-focused engineering effort.

He then went on to explain the two sides of his parallel approach to achieving AGI. The first part is what he called "Deep Machine Learning," a biologically-inspired framework for learning about the world with which we interact. A deep machine learning system builds a hierarchical model of the world in order to infer things about that world, and predict future events that might occur. It can discover structure based on spatial and temporal regularities in sensory observations, and deliver a powerful situation inference engine.

The complementary part is the decision making aspect of such a system, which he argues is driven by rewards. In particular, he feels reinforcement learning is at the core of intelligent decision making, which combined with deep machine learning can yield a breakthrough in AGI. Given the great advancements in large-scale integrated electronics, he claims that an AGI system which based on these principles may be built in the very near future.

Do you think we already have the building blocks to create AGI in the next ten or twenty years? Tell us about your thoughts right here on the blog.

 

 

 

 

 

 

 

 

 

 

6 Comments

  1. “The complimentary part is the decision”??? Who is paying a compliment? Is the writer trying to say “complementary?”

    “he claims that an AGI system which based on these principles…” How does an AGI _basde_ on these principles? Why is the writing here so sloppy?

  2. Because she’s live blogging? Don’t be such a drama queen.

  3. “_basde_”?

    It is very sad when one corrects another and yet cannot proofread one’s own correcting comment.

  4. “The complimentary part is the decision making aspect of such a system, which he argues is driven by rewards. In particular, he feels reinforcement learning is at the core of intelligent decision making, which combined with deep machine learning can yield a breakthrough in AGI.”

    Soooo…results based analysis instead of process based analysis? It seems to me that this should be rethought.

  5. both reinforcement learning and deep learning architectures are very much process driven, particularly since they are both analytically founded.

  6. With out going too deep into the subject let us see this a high level. Creating “Human Level” intelligence is the ultimate/ambitious target of AGI or any AI. Secondly “a biologically-inspired framework / Deep Learning architecture” when it comes to intelligence Human is the most successful biological evolution(I am not talking about running like a cheetah or climbing walls like a lizard purely in terms of decision making). What makes us humans so special? What the difference in the thinking process ? With going much into “philosophy of thought” we can say that be it be any situation in our life we have issues/problems and then we choose one of the available options. The MDP in RI in a way can be seen as a formal representation of this concept, and then with in our mind have feedback loops/lessons we learn and experiences/knowledge gain..

    Now coming to hierarchical structures, this is the way we humans operate we set priorities; priorities of different kinds information acquisition, actions to be performed etc..for example when I am in the class I would listen to the instructor rather than checking out the cricket scores in the web.!!!

    This for me seems quite an budding and interesting area of research which needs to be directed and worked out. Again I would to add this when it comes to humans the computational capacity and the intuition ability far is better by miles in comparison from the state of the art technology, so it at the a Human Level intelligent AGI tool was to be made the first steps are confine it to a small domain in terms of breadth on a reasonably defined problem.

    Lastly if I wrong in any thing or offended any one I am really sorry please take consider it as my ignorance and excuse me.